Handling Production-Data Uncertainty in History Matching: The Meren Reservoir Case Study
- C.S. Kabir (ChevronTexaco Overseas Petroleum) | N.J. Young (ChevronTexaco Overseas Petroleum)
- Document ID
- Society of Petroleum Engineers
- SPE Reservoir Evaluation & Engineering
- Publication Date
- April 2004
- Document Type
- Journal Paper
- 123 - 131
- 2004. Society of Petroleum Engineers
- 3.1 Artificial Lift Systems, 4.1.9 Heavy Oil Upgrading, 4.1.5 Processing Equipment, 2.4.3 Sand/Solids Control, 3 Production and Well Operations, 4.1.2 Separation and Treating, 5.3.2 Multiphase Flow, 5.1.5 Geologic Modeling, 3.3.1 Production Logging, 1.10 Drilling Equipment, 1.2.3 Rock properties, 5.1.8 Seismic Modelling, 5.2.1 Phase Behavior and PVT Measurements, 1.6 Drilling Operations, 3.1.6 Gas Lift, 4.3.4 Scale, 5.4.1 Waterflooding, 5.5 Reservoir Simulation, 5.1.2 Faults and Fracture Characterisation, 5.6.4 Drillstem/Well Testing, 5.1 Reservoir Characterisation, 3.3.6 Integrated Modeling, 5.5.8 History Matching, 5.5.7 Streamline Simulation
- 1 in the last 30 days
- 496 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 12.00|
|SPE Non-Member Price:||USD 35.00|
History matching reservoir performance ordinarily implies adjusting geologic and rock-fluid flow properties until a satisfactory match is attained between the observed and model responses. Property adjustments inevitably imply a trial-and-error approach.
Regardless of the approach used in any study, we presuppose the integrity of observed data. Unfortunately, measurements of gas and water do not have the desired precision when oil is the primary fluid of interest. For instance, when gas is flared, the quality of metering suffers. In some operations, withdrawal of small samples of a three-phase mixture near the wellhead becomes the sole basis for establishing water content in a production stream. These practices often introduce large uncertainty in the reported data of all three phases.
This paper explores how we dealt with some of these issues while attempting to match 30-year performance in the Meren field in offshore Nigeria. Results show that production data affected by wellbore mechanical problems can be diagnosed with log-log type curves. Furthermore, this work shows that discerning water cut from wellhead samples is prone to errors, and issues with wellbore flow calibration can lead to large uncertainties in performance prediction. Identification of various issues helped mitigate the risks, leading to successful infill wells as testified by their production performance.
Historically, reservoir simulation has been used to manage reservoirs. Key business decisions are often made on the basis of history-matched models. Some of these decisions include development drilling, facilities upgrade, workover schedules, stimulations, waterflooding, and installing artificial lifts. Of course, the quality of business decisions is driven largely by the quality of the history match.
One issue with any history-matching exercise is the quality of input data to the flow-simulation model. Geologic uncertainty is often cited as a major impediment to obtaining a quality history match on a well-by-well basis. Therefore, simulation engineers resort to parameter adjustments to mimic the observed response. Despite the availability of systematic approaches, such as stratigraphic1 and assisted-history-matching2,3 methods, individual subjectivity plays a large role while matching saturation [water cut or gas/oil ratio (GOR)] trends and production and injection profiles.
In history matching, the integrity of observed production data is rarely questioned; this is because business drivers seldom permit detailed investigations into every piece of simulator-input data. Nonetheless, the study of Pederson et al.4 stresses the need for validation of production data before initiating a simulation study.
In general, one or more input parameters, such as permeability (or its variation) and pore volume, is adjusted to obtain the desired match. But history matching, by definition, is an ill-posed inverse problem. In other words, the goodness of the match is no attestation to a model's robustness for forecasting purposes - the key ingredient for making business decisions. About a decade ago, Saleri5 shed light on the quality of performance prediction. Citing many in-house studies, he concluded that fieldwide prediction of cumulative oil production has an uncertainty spread of 10 to 40%. As expected, the error bar is even higher for individual wells, a result supported by Beliveau.6
We address three elements en route to our understanding of various data-related issues in history matching and forecasting. First, we explore uncertainty in water/oil ratio (WOR) and GOR data influencing history matching. Second, we discuss errors in water-cut measurements as often practiced in this and other fields. Third, we point out issues with well calibration that lead to uncertainty in forecasted results.
The Meren G-01/M-05 reservoir is contained within a northeast/southwest-trending, three-way fault-bounded structure, with a small amount of additional four-way closure at its crest, as Fig. 1 shows. The reservoir comprises two Miocene-age, shallow marine shoreface sand units (the G-01 and G-02 sands) that can be correlated fieldwide in Meren. The G-02 package is interpreted to be predominantly a progradational unit and consists of a succession of very clean, blocky mid- to upper-shoreface sands, with a number of relatively thin shale interbeds that represent minor marine flooding events. At the end of the G-02 deposition, a major marine flood took place as the shoreline receded, resulting in the deposition of the shale that lies between the G-02 and the overlying G-01 sand.
Maximum sand thicknesses occur in the M-05 well, where the G-01 and G-02 sands are 100 ft and 233 ft thick, respectively. Shale content within the G-01 and G-02 sands increases in a westward direction, representing overall reservoir degradation away from the main bounding/active faults.
The top G-01 structure map of Fig. 1 was constructed after the seismic horizon corresponding to the top of the G-01 sand interval (overlying the G-02) was interpreted on high-resolution 3D seismic data and then depth converted using a 3D velocity model with all available well and stacking velocity control. The top of the G-02 sand was assumed to be conformable to G-01; therefore, the G-02 depth map was made by isopaching down from G-01.
The G-01 and G-02 sands were subdivided into 16 total units based on correlation of the flooding surfaces within the reservoir. The interval between two flooding surfaces defines a region; each region was treated as one unit and independently simulated for property distribution.
Before upscaling, the earth-model grid had a dimension of 358×163×150, or 8,753,100 cells, with areal cell sizes of 75×75 ft and average thickness of 2 ft. These grid dimensions were selected to effectively model the reservoirs and maintain a maximum layer thickness of 3 ft in the fine-scale model.
|File Size||1 MB||Number of Pages||9|